One person's trash... The race to build more capable AI systems is pushing developers beyond the open web and into a far more intimate source of data: the internal workings of failed startups. As companies wind down, a growing secondary market has emerged for their digital exhaust – Slack threads, email chains, internal documents, and even source code.
What was once considered operational residue is now being packaged, scrubbed, and sold to AI developers seeking richer training environments. The shift reflects a broader evolution in how advanced AI models are built. Early large language models drew heavily from news archives, Wikipedia, and forums. Now, newer systems, particularly agentic AI, require something more structured and situational: data that mirrors how decisions unfold inside organizations.
To meet that need, developers are building "reinforcement learning gyms," controlled simulation environments where AI agents can rehearse workplace tasks. These systems rely on detailed, real-world datasets that capture workflows, communication patterns, and decision-making processes. The demand has become significant enough that Anthropic leaders have discussed spending up to $1 billion on such training infrastructure.
That demand is now intersecting with an unexpected supplier base – firms that specialize in shutting down startups. Companies like SimpleClosure, which typically handle payroll, taxes, and investor settlements during closures, are expanding into data monetization. Its newly launched platform, Asset Hub, is designed to help founders extract remaining value from their companies by licensing internal assets. These include not only technical materials, such as source code, but also workplace data, such as emails, documents, and Slack messages.
The company says it evaluates which data can be sold, estimates its value, and processes it to remove personally identifiable information before licensing. Forbes reports that over the past year, SimpleClosure has facilitated nearly 100 such transactions, with payouts ranging from $10,000 to $100,000 per company.
"There's a feeling of a gold rush from these companies trying to get their hands on real-world data," says SimpleClosure CEO Dori Yona.
Internal communications show how work actually happens – how teams coordinate, resolve ambiguity, and execute tasks. For AI systems designed to function as autonomous collaborators rather than passive tools, that context is difficult to replicate using public data alone.
However, the same qualities that make these datasets valuable also raise concerns. Unlike scraped web content, workplace communications often contain identifiable individuals, behavioral patterns, and sensitive exchanges. Even with anonymization, privacy advocates argue that the risks are not trivial.
"I think the privacy issues here are quite substantial," Center for AI and Digital Policy Founder Marc Rotenberg told Forbes. "Employee privacy remains a key concern, particularly because people have become so dependent on these new internal messaging tools like Slack … It's not generic data. It's identifiable people."
The concerns are beginning to draw attention from policymakers. The Center for AI and Digital Policy recently sent a letter to the Senate Commerce Committee urging the Federal Trade Commission to increase oversight of AI-driven businesses, particularly in how they source and use training data.
The trend links startup closures with AI development in a new way. As companies fail, their internal data – once ephemeral – can gain new utility as training material for the next generation of systems. In turn, those systems may reshape how future companies operate, communicate, and ultimately generate the very data that trains their successors.
The market is still developing but appears to be growing. The demand for more detailed, task-based data is increasing, while the supply – fueled by a steady churn of startup closures – shows little sign of slowing.